An Adaptive Oscillatory Neural Architecture for Controlling Behavior Based Robotic Systems

The introduction in Robotics of models inspired by biological clocks may be useful in order to cope with
a number of problems, like, for example, an efficient resources management in the sensorial pattern
elaboration, the coordination of different and parallel behaviors and the ability, for a robotic system, to
adapt its emergent behavior to different contexts providing an emergent action selection mechanism. In
this paper we present a general purpose neural-net able to obtain adaptive periodical controllers,
described by means of the NSBL. NSBL is a Neuro-Symbolic Behavior modeling Language that allows one
to express propositional logical inference and to translate them into the logically equivalent neural
network. Such general periodic clocks are peculiar to each behavior, and their periods are influenced by
the sensor input changing rate. In this way, the Robotic System is able to adapt its reaction time
coherently to the changes occurring in the environment and to its internal state. To test our architecture
we investigate the case of two conflicting behaviors.